Cognivo

• 

• 

• 

• 

• 

Inspiration:

What inspired me to create Cognivo was the desire to take language learning to the next level. When working with AI, you're always looking to solve real problems, and I identified a significant one in the language learning space.

The issue is that most people learn languages through apps like Duolingo or iTalki, which rely heavily on static lessons. This means learners are confined to predetermined content and can't really explore or practice beyond what each specific lesson covers. It's limiting and doesn't reflect how we naturally learn languages—through conversation and interaction.

Cognivo is designed to solve exactly that problem. Instead of static lessons, users learn languages at their own pace through dynamic, personalized conversations. The core concept is simple but powerful: users engage directly with conversational AI video tutors, creating a more natural and flexible learning experience that adapts to their individual needs and interests.

What it does:

Cognivo currently offers four languages: Spanish, French, Portuguese, and Japanese. Users can select one or multiple languages and progress through structured learning paths with multiple levels. Level 1 covers the fundamentals—alphabet, numbers, and basic building blocks. Level 2 introduces essential vocabulary like family members, household items, and everyday objects. By Level 3, users are engaging in conversations entirely in their target language.

The platform features two main sections: the Learning Path, which guides users through structured progression, and the AI Tutors section, which gives users direct access to every available tutor across all levels and languages.

Here's how it works: users choose their desired level and AI tutor, click 'Live Conversation,' then 'Join Call' to start their session. The user typically initiates the conversation, and the AI tutor responds based on the selected content. Let's say you choose Spanish Level 1 for the alphabet. You might start with a simple greeting, and the AI tutor will respond and ask if you're ready to learn the Spanish alphabet. Once you confirm, the tutor will begin with the letter 'A'—pronounced 'ah'—and provide a word that starts with that letter for you to practice and repeat.

The AI tutors actively correct mistakes, whether it's pronunciation, grammar, or vocabulary usage. If you want to switch topics—say from alphabet to numbers—you can simply end the call and select a different AI tutor.

What makes Cognivo especially powerful is that for every language, we have advanced AI tutors that conduct entire conversations in your target language. For Spanish, we have a restaurant simulation where you practice booking reservations, ordering food, asking questions, and requesting the bill—all in Spanish. It's like having real-world practice without leaving your home.

How I built it:

Now, let me walk you through the technical implementation. I built this using Bolt as my development platform and Tavus for the conversational AI video technology.

The setup process was quite manual but necessary for customization. For each AI tutor, I had to create them individually on the Tavus platform and carefully document three key pieces of information: the Persona ID, Replica ID, and Meeting Link. These identifiers were then fed into Bolt to integrate each tutor into the application.

Rather than using Tavus's default interface, I wanted to create a custom, branded experience that felt native to Cognivo. To achieve this, I implemented the Daily JS SDK integration method, which allowed me to build a fully customized interface for each Tavus Conversational Video Interface, or CVI.

The trickiest part was extracting the correct Meeting URLs. I had to manually start a call with each persona on the Tavus platform, copy the generated link, and then parse it down to get the essential URL format, which looks something like 'https://tavus.daily.co/' followed by a unique identifier.

This approach gave me complete control over the user experience while leveraging Tavus's powerful AI capabilities behind the scenes. It's more work upfront, but it ensures that users get a seamless, professional interface that matches Cognivo's brand and design language.

Challenges I ran into:

Throughout development, I encountered three major challenges, plus several minor ones that I'll touch on.

Challenge #1: Daily JS SDK Integration The biggest hurdle was implementing the Daily JS SDK integration method. Initially, I overlooked this approach and wasted considerable time trying alternative methods that simply weren't working. When I finally decided to explore the Daily JS SDK option, it turned out to be exactly what I needed. Combined with the difficulty of extracting the correct meeting links from Tavus, this challenge consumed a full 4 hours of troubleshooting.

Challenge #2: Subtitles and Transcripts I wanted to add subtitles and transcripts to each AI tutor session because the tutors sometimes speak too quickly, making them hard to understand. This feature would also enable users to take better notes during their learning sessions. Unfortunately, I ran out of Bolt tokens before completing this implementation, and even when I had sufficient tokens, I couldn't get it working properly. This remains an unresolved challenge.

Challenge #3: Custom Persona Creation We planned to let users create their own AI tutors—for interview preparation, custom conversation practice, or specialized scenarios. Any user-generated content would require administrator review to prevent malicious use. I successfully integrated the Tavus API for persona creation and fixed the initial bugs, but again ran out of tokens before finishing the feature. This is another challenge still on my to-do list.

Minor Challenges (All Resolved): I also faced several technical issues that I was able to overcome: microphone problems preventing users from speaking, webcam issues blocking video activation, audio problems where users couldn't hear the AI tutors, and call termination bugs where Tavus would still consider calls 'active' even after users ended them. Fortunately, all of these have been resolved.

Accomplishments that I'm proud of:

I'd definitely say the successful implementation of the Daily JS SDK integration method was our biggest achievement. Getting users to communicate with each AI tutor through a fully custom interface—that was the breakthrough moment for this entire project.

This feature took the longest to develop and caused the most frustration, but seeing it finally work was incredibly rewarding. It's what transforms Cognivo from just another language learning app into something that feels truly interactive and personalized. Without that custom interface, users would be stuck with generic video call layouts that don't match the learning experience we envisioned.

That integration is really the heart of what makes Cognivo special—it's the bridge between our educational content and the conversational AI technology that makes everything feel seamless and professional.

What I learned:

• How to integrate multiple APIs together (Bolt + Tavus + Supabase + Daily.js).
• Frontend UI/UX design for video call interfaces.
• Debugging integration issues between different platforms.
• Working with video/audio APIs and troubleshooting media permissions.
• Reading and implementing technical documentation
• Troubleshooting when multiple systems interact
• Learning when to pivot approaches (like switching to Daily JS SDK)
• Using Tavus.

What's next for Cognivo:

I have big plans for Cognivo's future. I'm working on implementing leaderboards to showcase top learners for each language, comprehensive achievement systems to keep users motivated, those subtitles and transcripts I mentioned earlier, the custom persona creation feature, structured learning paths, and much more.

This wasn't just a competition entry for me—Cognivo represents something I'm genuinely passionate about. I see real potential here to change how people learn languages, and I'm committed to continuing its development with the goal of eventually launching it as a full product for everyone to use.

I truly believe this project could make a meaningful impact in the language learning space. My primary goal is to secure the resources needed to bring Cognivo to market and build it into the comprehensive platform I envision. Even if I face resource constraints that prevent a full commercial launch, this project serves as a powerful proof of concept and portfolio piece that demonstrates what's possible when you combine conversational AI with personalized education.

Ultimately, I hope to get Cognivo out into the world where it can help real learners achieve fluency faster and more naturally. I'm also excited about the possibility of documenting this entire journey—creating a series that shows not just how I built it, but why I believe conversational AI is the future of language learning.

Built With

• bolt
• tavus